Humans are naturally quite good at controlling systems provided it can give input to the actuators via an interface (joystick etc.). In the humanoid community, motion primitives are obtained by logging human examples. Do you know any non humanoid research in which human input is logged to create motion primitives of the system it is controlling at that time?
Dear Hugo,
I suggest to you a links and attached files in topics.
The Artificial Intelligence Revolution: Part 2 - Wait But Why
waitbutwhy.com/2015/01/artificial-intelligence-revolution-2.html
-2015 : what do you think about machines that think? - Edge.org
https://www.edge.org/responses/q2015
-what scientific concept would improve everybody's toolkit? - Edge.org
https://www.edge.org/.../what-scientific-concept-would-improve-e...
-"If you can read this sentence, I can prove God exists"
cosmicfingerprints.com/prove-god-exists/
Best regards
Hello Hugo,
I propose you to have a look at our research concerning the identification of human motion primitives out of measured data in a vehicle driver steering scenario.
I believe human driver motion is an excellent application for any motion primitive related theory to proof concepts because here you can find highly trained movements performed nearly unconcious by the subjects and reduced degrees of freedom regarding the whole body due to the fixed sitting position. Additionally, active steering/braking/throttle systems are already available nowadays so the output from research is quite relevant.
Best regards
Conference Paper Individual Driver Modeling Via Optimal Selection of Steering...
Conference Paper An Identification Method for Individual Driver Steering Beha...
Conference Paper Experimental Validation of a Driver Steering Model Based on ...
Conference Paper Online Identification of Individual Driver Steering Behaviou...
Conference Paper Gray-Box Driver Modeling and Prediction: Benefits of Steerin...
Dear Hugo,
not sure whether it's exactly what you are asking for, but in our work we capture joystick input from an human operator, to "teach" four-legged robots how to perform specific tasks. This way, robot can learn from human experience, avoiding complex modelling of the environment and the robot dynamics. We use Case Based Reasoning (CBR) to map operator commands to robot response. CBR not only allows to achieve such mapping in an effective, simple way, but also the robot can develop its own version of the behavior from the learned one, adding its own experience parameters, to optimize the operator's performance.
Best regards
(You can find more information here:)
Article CBR based reactive behavior learning for the Memory-Predicti...
Thank you all for your answers and recommended reading!
I am considering a similar approach to you, Ignacio, to create a state space based mapping for an overactuated naval system.
- Badges
- Science topic
- Similar topics
- Engineering
- Control Theory
- Control Systems
- Control Systems Engineering